Deep Learning-Based Lower-Layer Upmixing

Páez T, Madrid L, Souza-Blanes E, Bharitkar S. Deep Learning-Based Lower-Layer Upmixing. In: Convention Paper. Audio Engineering Society; 2026. Paper 10283. Available from: https://aes.org/publications/elibrary-page/?id=23221

@inproceedings{Paez2026_23221,
  author = {Páez, Thaddeus and Madrid, Luis and Souza-Blanes, Ema and Bharitkar, Sunil},
  title = {{Deep Learning-Based Lower-Layer Upmixing}},
  note = {Paper 10283},
  year = {2026},
  month = may,
  publisher = {Audio Engineering Society},
  url = {https://aes.org/publications/elibrary-page/?id=23221}
}

TY  - CPAPER
TI  - Deep Learning-Based Lower-Layer Upmixing
AU  - Páez, Thaddeus
AU  - Madrid, Luis
AU  - Souza-Blanes, Ema
AU  - Bharitkar, Sunil
M1  - Paper 10283
PY  - 2026
DA  - 2026/05/28
UR  - https://aes.org/publications/elibrary-page/?id=23221
PB  - Audio Engineering Society
LA  - en
AB  - This paper introduces a novel approach for generating a lower layer in multichannel audio upmixing, addressing a gap in existing methods that primarily focus on mid and top layers. Leveraging Harmonic-Percussive Separation (HPS), the proposed framework dynamically adjusts key parameters (separation factor, harmonic attenuation, and phase shift) to enhance percussive components while diffusing harmonic elements. We compared three neural network architectures for this task: LSTM, TCN, and Transformer. Experimental results show comparable perceptual quality and objective metrics across all models, with the TCN being the most balanced and suitable for deployment on edge devices.
ER  -